A procedure for spectrophotometric statistical error calibration is described, in which estimates of the variance (sigma<sub>A</sub>2) in the absorbance A are derived as a function of A and the wavelength lambda, directly from recorded spectra. Spectra are obtained over the desired wavelength range for a number of neutral-density filters, and are then least-squares fitted piecemeal to low-order polynomials. The variances from these fits constitute the estimates of the instrumental variance sigma 2 for the lambda and A of the respective fitted segments. These values are in turn expressed as functions of lambda and A, by means of a small number of parameters. In the present case a representation in terms of A and I<sub>0</sub>, the 'effective' source intensity, is even more compact, requiring only five parameters to cover the 200-850nm region for absorbances up to and exceeding 2. For the instrument studied here, the high-A precision is limited by digitization error, the error being independent of lambda and I. Index Headings: UV-visible spectrophotometry; Statistical errors; Calibration; Least-squares weights; Spectral fitting.
Joel Tellinghuisen, "Statistical Error Calibration in UV-Visible Spectrophotometry," Appl. Spectrosc. 54, 431-437 (2000)